There are indeed examples of this out there in the wild; for example, VNC isn't launched as a separate process but rather as a thread in the target process with the Metasploit VNC shell option, IIRC.
Injecting code into other processes has plenty of use cases other than for hiding malicious code, however. (For example, VNC normally loads a DLL into processes being remoted in order to capture GUI-related activity. There are plenty of other examples of expected library injection.)
But this paper doesn't really relate all that much to Warden. It is more describing an approach whereby an existing project is reused (perhaps with a secondary thread or code hooks) to perform whatever approach is desired in terms of exploit code, instead of, say, starting a separate shell process for communication (which might be more obvious).
For what it's worth, though, Blizzard has not exactly had a stellar track record of security with respect to their network protocols. There have been plenty of bugs with the BNCS protocol in the past, exploitable both against the server and against clients to varying degrees, up to and including potential code execution due to bad message handling code.
